virtio_net: Make drop counter per-queue

Since when XDP was introduced, drop counter has been able to be updated
much more frequently than before, as XDP_DROP increments the counter.
Thus for performance analysis per-queue drop counter would be useful.

Also this avoids cache contention and race on updating the counter. It
is currently racy because napi handlers read-modify-write it without any
locks.

There are more counters in dev->stats that are racy, but I left them
per-device, because they are rarely updated and does not worth being
per-queue counters IMHO. To fix them we need atomic ops or some kind of
locks.

Signed-off-by: Toshiaki Makita <makita.toshiaki@lab.ntt.co.jp>
Signed-off-by: David S. Miller <davem@davemloft.net>

diff --git a/drivers/net/virtio_net.c b/drivers/net/virtio_net.c
index d03bfc4..7a47ce7 100644 (file)
--- a/drivers/net/virtio_net.c
+++ b/drivers/net/virtio_net.c
@@ -87,6 +87,7 @@ struct virtnet_sq_stats {
 struct virtnet_rq_stat_items {
        u64 packets;
        u64 bytes;
+       u64 drops;
 };
 
 struct virtnet_rq_stats {
@@ -109,6 +110,7 @@ static const struct virtnet_stat_desc virtnet_sq_stats_desc[] = {
 static const struct virtnet_stat_desc virtnet_rq_stats_desc[] = {
        { "packets",    VIRTNET_RQ_STAT(packets) },
        { "bytes",      VIRTNET_RQ_STAT(bytes) },
+       { "drops",      VIRTNET_RQ_STAT(drops) },
 };
 
 #define VIRTNET_SQ_STATS_LEN   ARRAY_SIZE(virtnet_sq_stats_desc)
@@ -705,7 +707,7 @@ err:
 
 err_xdp:
        rcu_read_unlock();
-       dev->stats.rx_dropped++;
+       stats->rx.drops++;
        put_page(page);
 xdp_xmit:
        return NULL;
@@ -728,7 +730,7 @@ static struct sk_buff *receive_big(struct net_device *dev,
        return skb;
 
 err:
-       dev->stats.rx_dropped++;
+       stats->rx.drops++;
        give_pages(rq, page);
        return NULL;
 }
@@ -952,7 +954,7 @@ err_skb:
                put_page(page);
        }
 err_buf:
-       dev->stats.rx_dropped++;
+       stats->rx.drops++;
        dev_kfree_skb(head_skb);
 xdp_xmit:
        return NULL;
@@ -1632,7 +1634,7 @@ static void virtnet_stats(struct net_device *dev,
        int i;
 
        for (i = 0; i < vi->max_queue_pairs; i++) {
-               u64 tpackets, tbytes, rpackets, rbytes;
+               u64 tpackets, tbytes, rpackets, rbytes, rdrops;
                struct receive_queue *rq = &vi->rq[i];
                struct send_queue *sq = &vi->sq[i];
 
@@ -1646,17 +1648,18 @@ static void virtnet_stats(struct net_device *dev,
                        start = u64_stats_fetch_begin_irq(&rq->stats.syncp);
                        rpackets = rq->stats.items.packets;
                        rbytes   = rq->stats.items.bytes;
+                       rdrops   = rq->stats.items.drops;
                } while (u64_stats_fetch_retry_irq(&rq->stats.syncp, start));
 
                tot->rx_packets += rpackets;
                tot->tx_packets += tpackets;
                tot->rx_bytes   += rbytes;
                tot->tx_bytes   += tbytes;
+               tot->rx_dropped += rdrops;
        }
 
        tot->tx_dropped = dev->stats.tx_dropped;
        tot->tx_fifo_errors = dev->stats.tx_fifo_errors;
-       tot->rx_dropped = dev->stats.rx_dropped;
        tot->rx_length_errors = dev->stats.rx_length_errors;
        tot->rx_frame_errors = dev->stats.rx_frame_errors;
 }